Disaggregated Computing System Accelerator Switching

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current large-scale computing architectures face inefficiencies in resource configuration and allocation, particularly in cloud computing, due to the limitations of fixed hardwired connections between processing elements and memory devices, leading to suboptimal data access and communication patterns that cannot scale economically or functionally.

Innovation Solution

The implementation of a disaggregated computing system that dynamically switches between different types of computing elements using general-purpose links and optical switching to create point-to-point connections, allowing for efficient communication and resource utilization by rewiring links between processors and memory elements based on workload demands.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If fixed hardwired connections are used between processing elements and memory devices, then system stability and reliability are improved, but adaptability and resource utilization deteriorate

Engineering Contradiction:

Improvesystem stabilityVSAvoidresource configuration flexibility

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent implements dynamic reconfigurability by allowing the system to switch between fixed hardwired connections and dynamically reconfigurable connections. Processing elements can be dynamically assigned to different memory devices based on workload demands, while maintaining stable operational states during execution. This resolves the contradiction by making the connection topology adaptable without compromising system reliability during operation.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent creates a universal interconnection fabric that can serve multiple functions: acting as fixed hardwired connections for stable operations and as dynamically reconfigurable pathways for adaptive resource allocation. The same physical infrastructure supports both reliability-critical fixed connections and flexibility-critical dynamic connections, eliminating the need for separate dedicated structures.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If dynamic reconfigurable connections are implemented, then adaptability and resource utilization are improved, but system complexity and implementation difficulty worsen

Engineering Contradiction:

Improveresource allocation flexibilityVSAvoidsystem implementation complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent introduces an intermediary resource allocation mechanism that manages the complexity of dynamic reconfiguration. This intermediary layer handles the switching and routing decisions, shielding the underlying complexity from both hardware implementation and software operation. The intermediary translates high-level resource allocation requests into specific connection configurations, reducing overall system complexity.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the reconfiguration control into manageable units, allowing different portions of the system to be reconfigured independently based on specific workload requirements. Rather than requiring system-wide reconfiguration, individual processing elements or memory devices can be dynamically allocated without affecting other segments, thereby reducing implementation complexity.

Inventive Principle:

Principle #1Segmentation

3Ease of operation

If traditional cloud computing resource allocation is used, then ease of service delivery is improved, but resource utilization efficiency and scalability worsen

Engineering Contradiction:

Improveservice delivery simplicityVSAvoidresource utilization efficiency

Core Design Contradiction:

Ease of operationVSProductivity

Solution Approach 1:

The patent implements feedback mechanisms that continuously monitor resource utilization and workload demands, automatically adjusting resource allocation to optimize efficiency. The system receives feedback on actual usage patterns and dynamically reconfigures connections to match demand, improving productivity while maintaining ease of operation through automated control. This eliminates the need for manual resource optimization while achieving superior utilization.

Inventive Principle:

Principle #23Feedback

Data Source

PatentUS11650849B2Efficient component communication through accelerator switching in disaggregated datacenters

Publication Date: 2023.05.16 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Embodiments are provided herein for efficient component communication and resource optimization in a disaggregated computing system. A first set of computing elements are used as in line accelerators and a second set of the computing elements are used as block accelerators within the disaggregated computing system. A switching operation is dynamically performed between the first set of computing elements and the second set of computing elements to perform a workload by rewiring one of a plurality of links associated with respective ones of the first set of computing elements and the second set of computing elements.